A polished cover with three incompatible scratch labels

Finishing inspector has agreed to join. They say the doubtful folder was someone's temporary holding area, then a later export treated it as a defect class. That explains one opinion.

Keep that export away from training data. Don't just rename the folder and assume nobody copied it. I'd tell the person building the dataset which labels are unverified.

Who owns the labels once the inspector decides?

Our quality lead will own the sample decisions. Imaging can propose a review, but not overwrite the decision to improve a trial result.

Yes. The moving band isn't a scratch, but there's an edge scuff. Inspector hasn't finished deciding whether that scuff fails this product's requirement.

Then it stays doubtful. Explaining the big stripe doesn't grant the little scuff a pardon.

I'd keep the scuff location with the images too, so the imaging person knows which physical feature the eventual decision concerns (not whichever line looks most dramatic).

Could the same cover be presented repeatably for that comparison?

There's a static nest from the tooling trial. We need to check that the cover seats consistently in it; until now people have held the samples by hand.

Show whoever loads it what seated looks like.

I'd still try a little deliberate presentation variation. A perfect nest photo won't tell you whether a minor normal seating difference makes the important scuff vanish.

Yes, once the nominal presentation is repeatable, have tooling and imaging define the variation to assess rather than letting everyone tilt it differently and call that the same trial.

Quality has rejected the scuffed cover. We've retained it with the scuff marked on its record. Static captures are being compared at the agreed presentation and nearby positions.

Can you still see the scuff in those nearby positions? That's the result I'd care about, not whether the big reflection behaves nicely.

Not reliably in the initial set. The images are repeatable, but one presentation makes the scuff very faint. So we have a better reference and still no accepted imaging setup.

Better to find that now.

Keep that faint view in the supplier comparison, with the approved rejection reason; otherwise somebody may select only the convincing pictures when presenting their proposed lighting.

Will the inspector review the next imaging trial too?